1. Deploy it in your own AWS or Azure
A small runner, installed with one Helm chart or the Terraform we ship for AWS, Azure and GCP, runs every app and agent inside your account. It connects outbound only: no inbound port, no VPN, no cloud credentials handed to us. It enrolls with a one-time token, is pinned to one namespace with least-privilege RBAC, and the images it pulls are signed and come with an SBOM. Updates reach it through the same tunnel; you can pause them.
Everything the platform does inside your cluster is written to an audit trail you can read in Settings or receive on a webhook, signed.
2. Keep AI inference within the EU
Each organization has an inference profile. Pick where calls to Anthropic and OpenAI models run: the platform default, your own API keys (and an EU gateway if you use one), Amazon Bedrock in an eu-* region, or Azure OpenAI in an EU region such as Sweden Central or France Central. Set the region to EU only and any call that could leave the EU is refused instead of sent, in the app, in the Inbox, and in the overnight runs alike.
IT tests the profile with one click before anyone uses it, and every call is logged with its provider and region, so "it stays in the EU" is something you verify, not something you are told.
3. Manage access, security and approvals centrally
- Sign-in through Microsoft Entra ID or Okta. Members are created, updated and removed by your directory over SCIM 2.0, and directory groups map to organization roles: admin, member, auditor. Ownership is never granted by a directory.
- A policy floor under every app. Which actions agents may never take, which always need a person, what amount counts as high risk, how many automatic actions a day. An app can only be stricter than the floor.
- Approvals decided by the right people. Organization admins see and decide every member's approvals. Above an amount you choose, two different admins must approve before anything runs, and it runs with the app owner's connections, never the approver's.
4. Give IT full visibility and control
One timeline per organization: sign-ins and MFA of every member, every change to the organization, every approval decision on every app, everything done in your clusters. Filter it, export it as CSV or NDJSON, or stream every event to your SIEM with an HMAC-SHA256 signature. Auditors read; admins act; nobody outside the organization sees any of it.
What is included
Enterprise
Everything in Business, plus the four promises above.
- Runs in your AWS, Azure or GCP account
- AI inference kept in the EU, verifiable per call
- SSO (Entra, Okta) and SCIM 2.0 provisioning
- Organization policies, two-admin approvals
- Audit trail with CSV / NDJSON export and SIEM webhook
- Named contact, security review support, yearly invoice
Priced per year on a quote, by number of members and apps. Pilots start on our cloud with the EU inference profile and SSO switched on, and move into your cloud when your IT is ready; nothing is rebuilt in between.
Questions IT asks
Does the runner need an inbound connection or our credentials?
No. The runner only connects outbound to the platform over TLS. It uses a one-time enrollment token and the permissions of the namespace you install it in. We never hold your cloud credentials.
Can we use our own model contracts?
Yes. Configure your Anthropic or OpenAI keys, your Bedrock account, or your Azure OpenAI resource in the organization's inference profile. Secrets are encrypted at rest and never shown again after entry.
What happens when someone leaves the company?
Your directory deactivates or deletes them over SCIM. Their membership is suspended at once; an account the directory created also loses its login. Their apps stay with the organization and its admins.
Where is our data?
On the enterprise plan, in your cloud account and region. On our cloud, EU customers' app databases are in Frankfurt or Berlin. The security page describes the controls in both cases.